Keep three layers separate
An observation describes the recorded evidence: for example, a dark linear feature visible across several camera frames. An interpretation proposes what it means: a possible surface discontinuity, a joint, staining or a shadow. A decision determines what to do: accept the present information, request a repeat view, commission qualified testing or initiate engineering assessment. Combining these layers in a single label such as “crack—repair required” can hide several unsupported steps.
Write findings so a reviewer can challenge the interpretation without losing the observation. Link each finding to original records, timestamps, locations and relevant quality notes. Use stable identifiers that survive a report revision. If an automated detector proposes a feature, retain that provenance and the human review outcome. An algorithmic confidence score is not automatically the probability that an asset has a particular defect.
Reconcile planned and usable coverage
Start with the agreed inspection zones and evidence criteria. Compare them with the actual sensor footprint and usable data, not only the vehicle track. A route may pass an area while the camera points elsewhere or the sonar view is shadowed. A high percentage of nominal coverage can conceal the one connection that matters most to the asset question. Report exclusions explicitly and explain their significance.
Where a coverage percentage is used, state its denominator and method. Is it surface area, length of pipeline, count of connections or number of planned views? Does it exclude inaccessible areas, and were those exclusions agreed before the mission? Avoid reporting percentages with precision that the underlying geometry cannot support. A clear map of observed, partly observed and unobserved zones often communicates more than a single aggregate value.
Preserve the measurement chain
Store original sensor records separately from processed products. Record software versions, processing settings, calibration files, mounting offsets and coordinate transformations where they affect the result. Derived mosaics, point clouds and meshes should be reproducible enough for the intended review. If processing fills gaps or smooths geometry, identify the operation and prevent the resulting surface from being mistaken for a direct observation.
Synchronisation deserves particular attention. A small time mismatch between a moving sensor and its navigation record can place a feature incorrectly. Check whether clocks were aligned, whether offsets changed during the mission and whether processing corrected them. State location uncertainty in terms useful for relocation. “Position available” is not the same as “this small feature can be found again with confidence.”
Make uncertainty operationally useful
Uncertainty has several sources: measurement, location, incomplete visibility and interpretation. These should not be compressed into one unqualified traffic-light label. Describe which uncertainty controls the next action. If geometry is well measured but the material surface is hidden by fouling, a more accurate navigation solution will not answer the remaining question. Targeted cleaning or a different measurement may be needed, subject to environmental and operational permissions.
“No defect detected” should be qualified by the method, coverage and demonstrated capability. It is not equivalent to proof that no defect exists. Repeated surveys introduce another issue: apparent change may come from different viewpoints, registration or environmental conditions. Compare like with like, quantify relevant differences and retain the possibility that an apparent trend is a measurement artefact until supported by adequate evidence.
An honest inspection record makes its blind spots easier to find than its best images.
Use the right engineering assessment
Inspection findings do not establish fitness for service on their own. For appropriate pipeline applications, DNV-RP-F101 addresses assessment of corroded pipelines, while DNV-RP-F105 concerns free-spanning pipelines. API 579-1/ASME FFS-1 provides fitness-for-service assessment methods within its scope. These references require suitable inputs and competent application; an attractive model or isolated thickness reading does not satisfy their assumptions automatically.
For underwater structural integrity, connect the review to the asset's established management system and applicable requirements rather than inventing a new acceptance rule in the survey report. Public guidance from organisations such as IMCA, the DNV standards catalogue and the ISO catalogue can help identify relevant documents. Check scope, current edition and contractual status before relying on any reference.
Close the loop with named actions
The final review should identify the responsible decision owner, accepted limitations, required follow-up and the trigger for reconsideration. A recommendation to “monitor” is incomplete unless it states what should be observed again, how comparable evidence will be obtained and what change would prompt action. Preserve disagreement between reviewers where it affects the outcome. A transparent exception is more useful than artificial consensus.